[llvm-dev] LLVM behavior different depending on function symbol name

[Mehdi AMINI via llvm-dev]

Hi,

2017-06-19 8:45 GMT-07:00 Andrew Kelley via llvm-dev <
llvm-dev at lists.llvm.org>:

> Greetings,
>
> I have a Zig implementation of ceil which is emitted into LLVM IR like
> this:
>
> ; Function Attrs: nobuiltin nounwind
> define internal fastcc float @ceil(float) unnamed_addr #3 !dbg !644 {
> Entry:
>   %x = alloca float, align 4
>   store float %0, float* %x
>   call void @llvm.dbg.declare(metadata float* %x, metadata !649, metadata
> !494), !dbg !651
>   %1 = load float, float* %x, !dbg !652
>   %2 = call fastcc float @ceil32(float %1) #8, !dbg !656
>   ret float %2, !dbg !657
> }
>
> Test case:
>
> test "math.ceil" {
>     assert(ceil(f32(0.0)) == ceil32(0.0));
>     assert(ceil(f64(0.0)) == ceil64(0.0));
> }
>
>
> When I compile with optimizations on, this test case fails. The optimized
> code for the test case ends up being a call to panic (assertion failure),
> which means that LLVM determined the test failed at compile-time.
>
> What's strange about this is that if I change the function name from @ceil
> to @ceil_asdf (and change the callers) then the test passes.
>
> So I think LLVM is doing some kind of string comparison on the symbol name
> and detecting that it is "ceil" and then having different, undesired
> behavior.
>
> I tried putting `nobuiltin` in the function attributes and at the
> callsite, but that did not change anything.
>
> Any ideas what's going on?
>

I think it'd be a lot easier to figure if you provide a standalone repro.

-- 
Mehdi
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20170619/e59ac2a9/attachment.html>